llvm dev - Oct 2019 - Wasm, start function, and default globals

Apologies if there is a better forum for these questions. Please redirect me if
so.

I’ve been using the clang/wasm-ld tools to experiment with some basic examples,
and there’s a couple things I’m wrestling with.

1) How to denote a function as the “start” function
(https://webassembly.github.io/spec/core/binary/modules.html#start-section)
2) How to avoid the defaulted __heap_base global.

I’ve dug around on the samples, mailing lists, and in the wasm-ld code for an
attribute or flag to specify the start function, but don’t see anything. Is this
just not implemented (or tracked) yet?

For 2), I always see the below global in the compiled output even with the most
trivial code that makes no reference to it (e.g. compiling just a simple
function that only references values on the implicit operand stack).

(global (;0;) (mut i32) (i32.const 66560))

Per https://dassur.ma/things/c-to-webassembly/, looks like this is the
__heap_base value implicitly provided for modeling the stack/heap as are usually
present in most C/C++ environments, even though unused in my compiled output. Is
this something that could/should be optimized away, but again just isn’t
implemented yet? Any way to suppress this building with the current
clang/wasm-ld toolchain?

Thanks,


  *   Bill


Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows
10

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191015/8af518db/attachment.html>

There is a good amount of expertise on this in Emscripten community: 
https://github.com/emscripten-core/emscripten

You can also try to ask in one of the WebAssembly repos: 
https://github.com/WebAssembly

I think Luke Imhoff has done some work accessing the start function, I 
am not sure if he is on this list. I personally don't know of a way 
write to start function. There is a bit on it in tool conventions:

https://github.com/WebAssembly/tool-conventions/blob/master/Linking.md#start-section

The __heap_base symbol is there because Wasm does not have the memory 
layout C family languages expect. At higher opt level Clang is usually 
pretty good at removing its use.

Best,

Petr

On 10/15/19 4:50 PM, Bill Ticehurst via llvm-dev wrote:
>
> Apologies if there is a better forum for these questions. Please 
> redirect me if so.
>
> I’ve been using the clang/wasm-ld tools to experiment with some basic 
> examples, and there’s a couple things I’m wrestling with.
>
> 1) How to denote a function as the “start” function 
> (https://webassembly.github.io/spec/core/binary/modules.html#start-section)
>
> 2) How to avoid the defaulted __heap_base global.
>
> I’ve dug around on the samples, mailing lists, and in the wasm-ld code 
> for an attribute or flag to specify the start function, but don’t see 
> anything. Is this just not implemented (or tracked) yet?
>
> For 2), I always see the below global in the compiled output even with 
> the most trivial code that makes no reference to it (e.g. compiling 
> just a simple function that only references values on the implicit 
> operand stack).
>
> (global (;0;) (mut i32) (i32.const 66560))
>
> Per https://dassur.ma/things/c-to-webassembly/, looks like this is the 
> __heap_base value implicitly provided for modeling the stack/heap as 
> are usually present in most C/C++ environments, even though unused in 
> my compiled output. Is this something that could/should be optimized 
> away, but again just isn’t implemented yet? Any way to suppress this 
> building with the current clang/wasm-ld toolchain?
>
> Thanks,
>
>   * Bill
>
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for 
> Windows 10
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191015/72585a9a/attachment.html>

The start function is not meant for general use, but rather for toolchains
to initialize the state of the runtime where they can guarantee that no
imports will be required.

For getting rid of unused globals and generally optimizing your wasm
modules, I highly recommend using Binaryen's wasm-opt tool.
https://github.com/webassembly/binaryen.

On Tue, Oct 15, 2019 at 6:11 PM Petr Penzin via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> There is a good amount of expertise on this in Emscripten community:
> https://github.com/emscripten-core/emscripten
>
> You can also try to ask in one of the WebAssembly repos:
> https://github.com/WebAssembly
>
> I think Luke Imhoff has done some work accessing the start function, I am
> not sure if he is on this list. I personally don't know of a way write
to
> start function. There is a bit on it in tool conventions:
>
>
>
https://github.com/WebAssembly/tool-conventions/blob/master/Linking.md#start-section
>
> The __heap_base symbol is there because Wasm does not have the memory
> layout C family languages expect. At higher opt level Clang is usually
> pretty good at removing its use.
>
> Best,
>
> Petr
> On 10/15/19 4:50 PM, Bill Ticehurst via llvm-dev wrote:
>
> Apologies if there is a better forum for these questions. Please redirect
> me if so.
>
>
>
> I’ve been using the clang/wasm-ld tools to experiment with some basic
> examples, and there’s a couple things I’m wrestling with.
>
>
>
> 1) How to denote a function as the “start” function (
> https://webassembly.github.io/spec/core/binary/modules.html#start-section)
>
> 2) How to avoid the defaulted __heap_base global.
>
>
>
> I’ve dug around on the samples, mailing lists, and in the wasm-ld code for
> an attribute or flag to specify the start function, but don’t see anything.
> Is this just not implemented (or tracked) yet?
>
>
>
> For 2), I always see the below global in the compiled output even with the
> most trivial code that makes no reference to it (e.g. compiling just a
> simple function that only references values on the implicit operand stack).
>
>
>
> (global (;0;) (mut i32) (i32.const 66560))
>
>
>
> Per https://dassur.ma/things/c-to-webassembly/, looks like this is the
> __heap_base value implicitly provided for modeling the stack/heap as are
> usually present in most C/C++ environments, even though unused in my
> compiled output. Is this something that could/should be optimized away, but
> again just isn’t implemented yet? Any way to suppress this building with
> the current clang/wasm-ld toolchain?
>
>
>
> Thanks,
>
>
>
>    - Bill
>
>
>
>
>
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for
> Windows 10
>
>
>
> _______________________________________________
> LLVM Developers mailing listllvm-dev at
lists.llvm.orghttps://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191016/7f8aa551/attachment-0001.html>

[sending this as a new email, as I didn’t get responses to my prior posting via
email – I just saw them on the archive]

Thomas Lively wrote via llvm-dev on Wed Oct 16 17:33:54 PDT 2019
> The start function is not meant for general use, but rather for toolchains
> to initialize the state of the runtime where they can guarantee that no
> imports will be required.
But I am trying to write a simple toolchain, hence trying to use it to
initialize my globals (or at least the one global I use for my memory
allocator).

Interestingly, if I have a global of a type with a simple constructor (e.g. just
setting members to integral values), I can see it doesn’t call the constructor,
but just inlines the resulting memory values. For example, if setting the first
member to 0xCCCC I see “(data (;0;) (i32.const 1024) "\cc\cc\00\00\etc..”
in the WAT. However if my global is a type with a non-trivial constructor, the
“data” entry for it is just zeroed out, and there is no startup function set to
run any initialization. (Which seems like a bug – or at least a compiler warning
should indicate globals will not be initialized as expected).

If this is something that SHOULD be added, happy to take a crack at it if
someone more experience in this area can provide a little guidance when
needed.
> For getting rid of unused globals and generally optimizing your wasm
> modules, I highly recommend using Binaryen's wasm-opt tool.
> https://github.com/webassembly/binaryen.
Yeah, just trying minimize the toolchain right now. Does Binaryen preserve the
“names” section? Clang/LLVM is populating that quite nicely, and some of the
stuff I’m experimenting with (e.g. profiling) depends on its presence.

Is the Binaryen or Emscripten or Wabt repo the best place to ask even Clang/LLVM
questions when it comes to Wasm? Seems most of the contributors are active on
those projects. (As the GitHub UX is better than this mailing list).

Thanks,


  *   Bill


Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows
10

From: Bill Ticehurst<mailto:billti at hotmail.com>
Sent: Tuesday, October 15, 2019 4:50 PM
To: llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>
Subject: Wasm, start function, and default globals

Apologies if there is a better forum for these questions. Please redirect me if
so.

I’ve been using the clang/wasm-ld tools to experiment with some basic examples,
and there’s a couple things I’m wrestling with.

1) How to denote a function as the “start” function
(https://webassembly.github.io/spec/core/binary/modules.html#start-section)
2) How to avoid the defaulted __heap_base global.

I’ve dug around on the samples, mailing lists, and in the wasm-ld code for an
attribute or flag to specify the start function, but don’t see anything. Is this
just not implemented (or tracked) yet?

For 2), I always see the below global in the compiled output even with the most
trivial code that makes no reference to it (e.g. compiling just a simple
function that only references values on the implicit operand stack).

(global (;0;) (mut i32) (i32.const 66560))

Per https://dassur.ma/things/c-to-webassembly/, looks like this is the
__heap_base value implicitly provided for modeling the stack/heap as are usually
present in most C/C++ environments, even though unused in my compiled output. Is
this something that could/should be optimized away, but again just isn’t
implemented yet? Any way to suppress this building with the current
clang/wasm-ld toolchain?

Thanks,


  *   Bill


Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows
10


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191017/972c26ae/attachment.html>

> Is the Binaryen or Emscripten or Wabt repo the best place to ask even
Clang/LLVM questions when it comes to Wasm?

The Emscripten repo is a nice catch-all for asking cross-cutting questions,
including wasm-specific Clang/LLVM questions. This list is of course fine
for LLVM questions as well.
> However if my global is a type with a non-trivial constructor, the “data”
entry for it is just zeroed out, and there is no startup function set to
run any initialization.

It sounds like you're looking for __wasm_call_ctors. I was going to link to
its documentation, but it turns out there is none. I'll fix that here:
https://github.com/WebAssembly/tool-conventions/blob/master/Linking.md.
Anyway, just add that function to the list of functions lld should export
and make sure to call it in your runtime. You're not seeing your
constructor functions now because they're being garbage collected by lld.
> Does Binaryen preserve the “names” section? Clang/LLVM is populating that
quite nicely, and some of the stuff I’m experimenting with (e.g. profiling)
depends on its presence.

Yes, Binaryen can preserve the names section if you pass it -g.

On Thu, Oct 17, 2019 at 12:58 PM Bill Ticehurst via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> [sending this as a new email, as I didn’t get responses to my prior
> posting via email – I just saw them on the archive]
>
>
>
> Thomas Lively wrote via llvm-dev on Wed Oct 16 17:33:54 PDT 2019
>
>
>
> > The start function is not meant for general use, but rather for
> toolchains
>
> > to initialize the state of the runtime where they can guarantee that
no
>
> > imports will be required.
>
>
>
> But I am trying to write a simple toolchain, hence trying to use it to
> initialize my globals (or at least the one global I use for my memory
> allocator).
>
>
>
> Interestingly, if I have a global of a type with a simple constructor
> (e.g. just setting members to integral values), I can see it doesn’t call
> the constructor, but just inlines the resulting memory values. For example,
> if setting the first member to 0xCCCC I see “(data (;0;) (i32.const 1024)
> "\cc\cc\00\00\etc..” in the WAT. However if my global is a type with a
> non-trivial constructor, the “data” entry for it is just zeroed out, and
> there is no startup function set to run any initialization. (Which seems
> like a bug – or at least a compiler warning should indicate globals will
> not be initialized as expected).
>
>
>
> If this is something that SHOULD be added, happy to take a crack at it if
> someone more experience in this area can provide a little guidance when
> needed.
>
> > For getting rid of unused globals and generally optimizing your wasm
>
> > modules, I highly recommend using Binaryen's wasm-opt tool.
>
> > https://github.com/webassembly/binaryen.
>
>
>
> Yeah, just trying minimize the toolchain right now. Does Binaryen preserve
> the “names” section? Clang/LLVM is populating that quite nicely, and some
> of the stuff I’m experimenting with (e.g. profiling) depends on its
> presence.
>
>
>
> Is the Binaryen or Emscripten or Wabt repo the best place to ask even
> Clang/LLVM questions when it comes to Wasm? Seems most of the contributors
> are active on those projects. (As the GitHub UX is better than this mailing
> list).
>
>
>
> Thanks,
>
>
>
>    - Bill
>
>
>
>
>
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for
> Windows 10
>
>
>
> *From: *Bill Ticehurst <billti at hotmail.com>
> *Sent: *Tuesday, October 15, 2019 4:50 PM
> *To: *llvm-dev at lists.llvm.org
> *Subject: *Wasm, start function, and default globals
>
>
>
> Apologies if there is a better forum for these questions. Please redirect
> me if so.
>
>
>
> I’ve been using the clang/wasm-ld tools to experiment with some basic
> examples, and there’s a couple things I’m wrestling with.
>
>
>
> 1) How to denote a function as the “start” function (
> https://webassembly.github.io/spec/core/binary/modules.html#start-section)
>
> 2) How to avoid the defaulted __heap_base global.
>
>
>
> I’ve dug around on the samples, mailing lists, and in the wasm-ld code for
> an attribute or flag to specify the start function, but don’t see anything.
> Is this just not implemented (or tracked) yet?
>
>
>
> For 2), I always see the below global in the compiled output even with the
> most trivial code that makes no reference to it (e.g. compiling just a
> simple function that only references values on the implicit operand stack).
>
>
>
> (global (;0;) (mut i32) (i32.const 66560))
>
>
>
> Per https://dassur.ma/things/c-to-webassembly/, looks like this is the
> __heap_base value implicitly provided for modeling the stack/heap as are
> usually present in most C/C++ environments, even though unused in my
> compiled output. Is this something that could/should be optimized away, but
> again just isn’t implemented yet? Any way to suppress this building with
> the current clang/wasm-ld toolchain?
>
>
>
> Thanks,
>
>
>
>    - Bill
>
>
>
>
>
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for
> Windows 10
>
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191017/ce160291/attachment.html>